Overclock.net banner
Cancel
Create thread New Forums

AMD DDR5 OC And 24/7 Daily Memory Stability Thread

93 reading
Jump to Latest
3,281 - 3,300 of 31,499 Posts
#3,281 ·
babbalucio said:
Well I did a rapid test at SOC 1.15V with superPI 1M, same exact timings, test repeated few times to check consistency

2000fclk -> 6.014-6.075 secs
2133fclk -> 5.993-6.055 secs

I don't see regressions, it gets slightly better. Maybe 1.15V is enough? or should I try with higher soc voltage?
Click to expand...
Lucio best method to capture fclk instability is LATENCYMON + REALBENCH, all my benchmarks were better at 2133 against 2100 but something strange happen (at least with my actual voltages) at 2133 because latencymon goes RED. At 2100/6000 i am on the edge of full stability using vsoc 1.15v.
Generally high process latency was captured inside Latencymon within 15 minutes after Realbench.
 
#3,282 ·
  • Haha
Reactions: unsivilaudio
Save
Reply Quote Rep+ Rep+
#3,283 ·
Alexgti79 said:
Lucio best method to capture fclk instability is LATENCYMON + REALBENCH, all my benchmarks were better at 2133 against 2100 but something strange happen (at least with my actual voltages) at 2133 because latencymon goes RED. At 2100/6000 i am on the edge of full stability using vsoc 1.15v.
Generally high process latency was captured inside Latencymon within 15 minutes after Realbench.
Click to expand...
aida64 latency benchmark also very fast to display unstable error correcting fclk
 
#3,284 ·
Alexgti79 said:
Lucio best method to capture fclk instability is LATENCYMON + REALBENCH, all my benchmarks were better at 2133 against 2100 but something strange happen (at least with my actual voltages) at 2133 because latencymon goes RED. At 2100/6000 i am on the edge of full stability using vsoc 1.15v.
Generally high process latency was captured inside Latencymon within 15 minutes after Realbench.
Click to expand...
high as in constantly high? or just high peak?
 
#3,286 ·
Sooooo just bought a 7900x along with a ASUS TUF GAMING PLUS B650 + Kingston fury beast 5600Cl36 2x16 ram which is most likely Hynix M-die, Are there any good baselines settings to run to get at, or near 60ns mem latency? Are the Settings found here (223) Easy memory timings for Hynix DDR5 with Ryzen 7000 - YouTube a good baseline, as i think i read somewhere that the timings mentioned in this video is somewhat wrong and slows down the ram by a little.

Also with the current issues of burned CPU's should i limit soc to max 1.2-1.25v?
 
#3,287 ·
Yviena said:
Sooooo just bought a 7900x along with a ASUS TUF GAMING PLUS B650 + Kingston fury beast 5600Cl36 2x16 ram which is most likely Hynix M-die, Are there any good baselines settings to run to get at, or near 60ns mem latency? Are the Settings found here (223) Easy memory timings for Hynix DDR5 with Ryzen 7000 - YouTube a good baseline, as i think i read somewhere that the timings mentioned in this video is somewhat wrong and slows down the ram by a little.

Also with the current issues of burned CPU's should i limit soc to max 1.2-1.25v?
Click to expand...
Try tRFC 370 @5600. If it works, it is Hynix A die.
 
#3,288 ·
babbalucio said:
I'm wondering if 1.15V soc is a bit too low for my config, maybe 1.2 is a more "safe" value?
I've been using 1.15V for almost two weeks with any apparent problems at all and very intensive workloads (video editing, reandering, gaming, multiplayer)
What should I look at to understand if soc voltage is too low?
you say consistency in superPI, I get all the 1M tests in a range of 50ms
all the Pyprime in a range of 50ms too, with the exception of the first try maybe

Image
Click to expand...
babbalucio said:
Well I did a rapid test at SOC 1.15V with superPI 1M, same exact timings, test repeated few times to check consistency

2000fclk -> 6.014-6.075 secs
2133fclk -> 5.993-6.055 secs

I don't see regressions, it gets slightly better. Maybe 1.15V is enough? or should I try with higher soc voltage?
Click to expand...
Test the Aida Sha3 benchmark. That one usually pretty hard crash if something is unstable.
 
#3,289 · (Edited)
donnieyeen said:
Test the Aida Sha3 benchmark. That one usually pretty hard crash if something is unstable.
Click to expand...
not a single crash in Sha3 aida test :cool:

10 consecutive tests with 1.15V soc at 6000C30 and 2133fclk (I also have a -15 CO all cores, just the undervolt without any fmax override)

cpu temp peaked around 86 degrees, memory under 40, every result around 11000 MB/s (it's a 7950x3d)

rock solid at 1.15V I suppose, I don't see any reasons to go higher with soc voltage after all (y)

EDIT: I noticed something else during the Sha3 testing, the SOC voltage spiked to 1.162V, that's a 12mV difference from what is set in bios.
Not an Asus :LOL: but better pay attention at those fluctuations.

Alexgti79 said:
Lucio best method to capture fclk instability is LATENCYMON + REALBENCH, all my benchmarks were better at 2133 against 2100 but something strange happen (at least with my actual voltages) at 2133 because latencymon goes RED. At 2100/6000 i am on the edge of full stability using vsoc 1.15v.
Generally high process latency was captured inside Latencymon within 15 minutes after Realbench.
Click to expand...
I find latencymon too much driver dependent.
For example, if you read latest known issues in nvidia drivers you find "Increased latency in latencymon"
And in fact nvlddmkm.sys goes crazy high very often.
The solution is to enable MSI for the geforce card and set the affinity to a core different from core 0 (core 6 for example)
That solves the high spikes in latencymon. Still can't understand why 4000rtx cards don't use MSI by default... but that's OT
 
#3,290 ·
Hairy MacChoir said:
I am happy with mine done without DDR5 high voltage mode

View attachment 2614681
Click to expand...
Fresh install with EXPO I activated (full auto then), and my results are way worse, despite higher voltages mostly everywhere ! I guess I will take inspiration from your settings
Do you happen to have your bios settings handy somewhere ? (I appreciate ZenTimings give some of those settings :p )
 
#3,291 ·
latency mon doesn't go above 270ns for me when playing warzone 2 with a 7800x3d and 4090 but if pc is idle and i open discord it can jump to 1000ns and seems like Nvidia isn't going to fix it because they had that as a known issue for very long time now
 
#3,294 ·
dansi said:
aida64 latency benchmark also very fast to display unstable error correcting fclk
Click to expand...
in my case aida is always perfect every run, all is fine and good. Problems appear in heavy load situation after some minutes, in games appear like stuttering everywhere
dnd1 said:
high as in constantly high? or just high peak?
Click to expand...
When everything is stable you never see high process latency, not even a single peak! Latencymon must stay green all the time
 
#3,295 ·
I'm so tired of these. all my overclocks were stable at that point, a few days later everything was not stable anymore. I had to adjust so many times. It's stable for several days with no crashes during stress tests or gaming. But a few days later it became unstable again. Do i have a bad cpu or is my ram too bad, or is it because the motherboard with the bios is unstable?
 
Save
Reply Quote Rep+ Rep+
#3,296 ·
Alexgti79 said:
in my case aida is always perfect every run, all is fine and good. Problems appear in heavy load situation after some minutes, in games appear like stuttering everywhere
When everything is stable you never see high process latency, not even a single peak! Latencymon must stay green all the time
Click to expand...
Latencymon is useless,.. the Nvidia drivers for example have a bug as somebody mentioned which increases the latency no matter. There is no way you are staying green on latest Nvidia drivers. They also haven't fixed it since several revisions and have that in changelog.
 
#3,297 · (Edited)
PC: R5 7600x, B650e-e bios 1409 and KINGSTON RGB 6400 CL32 2x16 KIT A-DIE i think 03/2023 production HW-INFO reports this kit as A-DIE

CURRENT AIDA SCORES Read +-70100 Write +-96600 Copy +-69900 Latency 53.3-54.1 NS Cinebench r23 Muliti +- 15200-300

6200 MHZ 28-38-38-28 FLCK 2200 GDM DISABLED POWER DOWN DISABLED

Cpu SoC Manual 1.16 (bios reports 1.176)
VDD Voltage 1.46 VDDQ Voltage 1.46
VDDIO/MC Voltage 1.33 in bios (bios reports 1.36)

TRC 56 (i think even 48 possible becouse 35 boots but with errors) Twr 48 Refresh interval 65535 Trfc1 450 Trfc2 350 Trfcsb 250 Trtp 12 TrrdL 8 TrrdS 4 Tfaw 20 TwtrL 14 TwtrS 4 Trdrd ScL 6 Trdrd SC 1 Trdrd Sd AUTO Trdrdddd AUTO Twrwr SCL 6 Twrwr LC 1 Twrwr Sd AUTO Twrwr Dd AUTO Twrrd 2 Trdwr 14

this settings tested 250 cyctes usmuss v3 12+ hours anta777extreme 12+hours y-cruncher VST and OCCT AVX large

TIPS for turning GDM OFF for me it was ScL timmings need to go from 4 to at least to 5 i stayed at 6 but i reccomend to start from 8 I also heard that too low Tfaw is a problem too so i reccomend 24 for beginning i got mine at 20 as i said before ScL was the couse in my situation. with gdm on i got easy scl both timmings at 4 but worse latency average +-1NS

6400 cl 32 stock timmings working without problems flck 2200 but firstly i got problems when i start overclocking memory and i decided to tune max for 6200 MHZ so in near future i try max 6400 i think 6400 CL 30 FLCK 2200 highly possible with GDM enabled and a bit worse timmings

first guys stabilize your cpus ! thats was the biggest misleading problem and couse of most errors !
i think i got pretty bad cpu oc potencial and max boost clock set to 5.4 GHZ 1.305 max voltage but good memory controller on this unit another way ...
i had to set -50 max boost clock in bios and positive all cores curve optimizer to value 8
i was sure cuz everything cpu voltage related is on auto it have to be stable but it definitly wasnt !
 
#3,298 ·
MSIMAX said:
Drag Racing

View attachment 2614875
Click to expand...
nice result, can you test 6500 mhz bclk oc ?? I can do this on my system , but soc 1,37(cold)1,395(hot) neded. i can easy go 6490mhz @ soc 1,265-1,270 mv.

please try frequency 6500 with timings cl26-36-36-30-68-378 @ 1,64-1,68v

I think it's best if you don't want to SOC above 1.3v for regular 24/7 use.
 
3,281 - 3,300 of 31,499 Posts
About this Discussion
31K Replies
1.1K Participants
Last post:
gupsterg
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling